Object Detection on Unmanned Arial Vehicles Dataset Using Adaptive HydraNet
| dc.contributor.advisor | Leung, Henry | |
| dc.contributor.author | Naseri Golestani, Sara | |
| dc.contributor.committeemember | Behjat, Laleh | |
| dc.contributor.committeemember | Ioannou, Yani | |
| dc.date | 2023-06 | |
| dc.date.accessioned | 2023-05-09T16:46:10Z | |
| dc.date.available | 2023-05-09T16:46:10Z | |
| dc.date.issued | 2023-04-18 | |
| dc.description.abstract | Recent years have witnessed substantial developments in object detection methods. However, detecting medium and small objects on Unmanned Aerial Vehicle (UAV) datasets remains a significant challenge due to the limitations of the current backbone architecture of these methods. This limitation arises from the architecture's multilabel classification step, which lacks precision in detecting small objects and consumes large amounts of computational resources. This study proposes a novel solution to overcome this limitation by introducing AHydraNet, a multitask learning module based on the low-cost dynamic multitask architecture HydraNet. AHydraNet is a multilabel classification template with an adaptive threshold that enhances the precision of the detection for small and medium-sized objects. We integrate AHydraNet into the Mask R-CNN's backbone by introducing a smaller module called the Adaptive Branching Network (ABN), which applies AHydraNet to all the output feature maps of the feature pyramid network. The resulting model is called AHydraFPN. The performance of AHydraFPN is evaluated on two popular datasets, MS-COCO and Arial-Cars, and compare it with the performance of Mask R-CNN. Our experimental results demonstrate that AHydraFPN achieves a significant improvement on average recall (AR) than the baseline model. These results indicate that our proposed solution can remarkably improve the detection of small and medium-sized objects on UAV datasets. | |
| dc.identifier.citation | Naseri Golestani, S. (2023). Object detection on unmanned arial vehicles dataset using adaptive HydraNet (Master's thesis, University of Calgary, Calgary, Canada). Retrieved from https://prism.ucalgary.ca. | |
| dc.identifier.uri | http://hdl.handle.net/1880/116194 | |
| dc.identifier.uri | https://dx.doi.org/10.11575/PRISM/dspace/41039 | |
| dc.language.iso | en | |
| dc.publisher.faculty | Graduate Studies | |
| dc.publisher.institution | University of Calgary | |
| dc.rights | University of Calgary graduate students retain copyright ownership and moral rights for their thesis. You may use this material in any way that is permitted by the Copyright Act or through licensing that has been assigned to the document. For uses that are not allowable under copyright legislation or licensing, you are required to seek permission. | |
| dc.subject | Multitask Learning | |
| dc.subject | Object Detection | |
| dc.subject | Multilabel Classification | |
| dc.subject | UAV Datasets | |
| dc.subject.classification | Artificial Intelligence | |
| dc.title | Object Detection on Unmanned Arial Vehicles Dataset Using Adaptive HydraNet | |
| dc.type | master thesis | |
| thesis.degree.discipline | Engineering – Electrical & Computer | |
| thesis.degree.grantor | University of Calgary | |
| thesis.degree.name | Master of Science (MSc) | |
| ucalgary.thesis.accesssetbystudent | I do not require a thesis withhold – my thesis will have open access and can be viewed and downloaded publicly as soon as possible. |